SENTINEL (logo)
GIF
Scrolling
GIF
Levantar reportes
Conexión con API
Implementación de backend
GIF
Desglose de compras hormiga

Sentinel Banca Inteligente

💡 Inspiration

Our journey began with a single, staggering statistic: Capital One faces a potential loss of up to $400 million annually, not from the fraud itself, but from loyal customers leaving after a negative resolution experience. A fraud event is the ultimate "moment of truth" for a bank, yet the current process often adds anxiety and frustration to an already stressful situation.

Data shows that poor service is the #1 reason for customer churn in banking, and 56% of customers who leave say the bank could have done something to keep them.[1]

We were inspired to transform this critical point of friction from a liability into a loyalty-building asset. We asked ourselves: What if we could turn a customer's worst day into the very reason they trust us for life?

🚀 What it does

Sentinel is an intelligent shield integrated directly into the Capital One app, designed to protect, empower, and reassure customers.

Proactive AI Detection:
Using a sophisticated Machine Learning model, Sentinel analyzes every transaction in real-time. It flags anomalies by looking at patterns in location, merchant history, and time of purchase. This powers our smart, color-coded notification system:
- 🔵 Blue: All clear! A normal, everyday purchase.
- 🟠 Orange: A "ghost charge" alert. Sentinel spots small, recurring charges like forgotten subscriptions, making it easy to review and cancel them.
- 🔴 Red: High-risk transaction detected. The user gets an instant, actionable alert to verify the purchase or report fraud with a single tap.
Transparent Resolution Portal:
When fraud is reported, the black box of uncertainty disappears. Sentinel provides a dedicated in-app command center where the customer has total control:
- 🗺️ Visual Case Roadmap: Track the claim's progress from Recibido → En Revisión → Resuelto, just like tracking a package.
- ❓ Interactive FAQs: Get instant answers to common questions, reducing the need to call support.
- 💬 Instant Support: Connect immediately with a human agent with one click.

🛠️ How we built it

We built Sentinel with a modern, robust tech stack focused on performance, security, and a seamless user experience.

Frontend (UX):
- TypeScript & React: We chose this combination to build a clean, responsive, and intuitive user interface. Our design philosophy was to create a calming and empowering experience for a user in a high-stress situation.
Backend & Machine Learning (AI):
- Python: The engine of our detection system is built in Python, the industry standard for data science.
- Isolation Forest: We implemented this unsupervised learning algorithm because it's exceptionally good at anomaly detection—perfect for identifying the "needle in a haystack" that is a fraudulent transaction among millions of legitimate ones.[2]

-DevOps & Deploy Render (Backend) GitHub Pages (Frontend)

Data (API):
- Capital One API: We leveraged the official API to simulate a real-time stream of transactional data, allowing us to train and validate our model's ability to distinguish between normal and suspicious behavior.

🧗 Challenges we ran into

Balancing Model Sensitivity: Tuning the AI was a tightrope walk. Making it too sensitive would create "alert fatigue" by flagging legitimate purchases. Too lenient, and it would miss real fraud. Finding that perfect balance required meticulous feature engineering and threshold adjustments.
Designing for Clarity, Not Panic: The UI for a fraud alert is critical. Our initial designs for the red alert were perceived as too alarming. We iterated extensively on the design, colors, and copy to be direct and informative without causing panic, empowering the user to act calmly.
Simulating Realistic Fraud: Creating a synthetic dataset that accurately mimics the diversity of real-world transactions and includes subtle, sophisticated fraud patterns was a significant challenge.

🏆 Accomplishments that we're proud of

An Effective & Efficient AI Model: Successfully implemented an Isolation Forest model that demonstrated a high accuracy rate in identifying simulated fraud while keeping false positives to a minimum.
An Empowering User Experience: The user-centric resolution portal, especially the visual roadmap, was consistently praised in user testing for reducing anxiety.
A Rock-Solid Business Case: Built a financial model quantifying the problem and massive ROI, projecting the retention of over $100 million in annual revenue.
A Fully Functional Prototype: Created a working, end-to-end prototype showcasing the entire user journey. -Our first full-stack proyect We feel proud of what we achieved on this hack

🧠 What we learned

Transparency is the New Security: Reducing customer anxiety through clear communication is as important as the underlying technology.
AI as a UX Tool: AI/ML can create hyper-personalized and proactive user experiences that build deep, lasting trust.[2]
The Economics of Empathy: Solving a critical customer pain point can have an outsized economic impact, unlocking a significant business opportunity.